Counting insects in flight using image processing techniques

Abstract

A method to monitor New Zealand's native bees using image processing technology is presented. Since most species are solitary ground nesting bees the number of active nests within an area can give a good estimate of the population of a community. The number of native bees in flight around plants can also provide valuable information about the overall health of a community and help to quantify their value in the ecosystem as keystone pollinators. On this basis, images of insects in flight have been collected across one season and the results compared with previous results of active nest counts. Open source software FIJI was used to pre-process and classify images. Accuracies were verified using data mining software WEKA. Performance evaluations showed the fast random forest classifier consistently returned fast, accurate results. Fine differences in images were discriminated that were otherwise impossible to identify with the naked eye and even when training data were unevenly distributed the classifier returned accuracies above 98%. The results are promising and while there are few alternatives to traditional methods, image processing for ecology can provide cost effective, standardized tools to help monitor the population and diversity of native bees in New Zealand.